Introduction

Osteoporosis is characterized by low bone mass and microstructural deterioration leading to an increased risk of fracture. Bone mass and, with certain limitations, structure can both be assessed with bone densitometric imaging techniques in vivo. Areal bone mineral density (BMD) as assessed by dual X-ray absorptiometry (DXA) remains the prevailing bone densitometry parameter used in the clinical setting. However, DXA has limitations in its ability to measure the mass distribution in cortical and trabecular bone compartments and in the evaluation of bone geometry and microstructure. This is the domain of quantitative computed tomography (QCT).

Single slice QCT, the foundation of all QCT techniques discussed in this review, was introduced over 3 decades ago [1–3] to assess apparent bone mineral density (BMD) in the trabecular center of the vertebrae and in the distal forearm [4]. Since then, technical advances in clinical CT-technology have led to continuous improvements in QCT applications. In the 1990s, the introduction of spiral CT allowed for 3D acquisition and made possible refined 3D analysis of volumetric data [5]. This in turn extended QCT imaging to the hip. Improved detector technology combined with cone beam acquisition originally developed for μCT opened the door for high resolution peripheral QCT (hr-pQCT) to assess bone architecture in vivo. Finally finite element analysis (FEA), a method developed in mechanical engineering and first applied in an anatomy-specific manner to bone in 1989 [6] is now a mature, validated technique based on 3D CT datasets that integrates BMD with bone geometry and structure to estimate bone strength under various loading conditions [7].

The increasing clinical use and interest in CT-based techniques in the field of osteoporosis is triggered by increasing evidence that measuring not only bone mass and density but also structure and strength are relevant to the estimation of fracture risk and the understanding of potentially differential therapeutic effects. We provide state of the art information on human applications of CT in the field of osteoporosis.

Measuring Mass and Density

QCT is a volumetric technique, (ie, it provides a separate measure of BMD for each voxel of the scanned volume of interest [VOI]). In particular, cortical and the metabolically more active trabecular compartments can be measured separately (Fig. 1). BMD in g/cm3 is calculated from the CT attenuation values by means of a calibration phantom scanned with the patient. The phantom contains different concentrations of hydroxyapatite equivalent material. Bone mineral content (BMC) is the product of BMD and volume. QCT T-scores, analogous to those derived from DXA BMD, can be calculated using appropriate normal reference data but their relevance to clinical care is limited by the considerable discordance from DXA T-scores [8, 9]. DXA T-scores remain the sole basis for the accepted WHO diagnostic criterion for osteoporosis. An adaptation, in which DXA equivalent areal BMD but not true volumetric BMD results are obtained from a 3D QCT dataset (CTXA method; Mindways Inc, Austin, TX) overcomes that limitation of the clinical application of QCT [10]. While many cross-sectional studies have shown the association of QCT with prevalent fracture, so far only a few studies have demonstrated an association between QCT BMD and incident fracture. In the EFFECT study and a similar study conducted in China, both of which included patients with fresh femur fractures [11•, 12], QCT of the hip improved hip fracture discrimination at best marginally when compared with DXA. Similar results were observed in prospective studies of incident hip fracture including the AGES-REYKJAVIK study [13•] and the MrOS studies [14, 15]. In contrast a recent study showed that at the spine, fracture risk prediction was improved by QCT imaging [16••] supporting results from a number of earlier cross-sectional studies [9, 17–20].

Fig. 1
figure 1

QCT of the spine (top) and hip (bottom). Top left: spine segment L1-L3 with intervertebral spaces. Top center: sagittal MPR with periosteal (red) and cortical and subcortical (blue) contours; superior and inferior VOIs are separated. Top right: axial MPR with traditional central elliptical and Siemens Osteo VOIs. Bottom left: analysis VOIs of the proximal femur: head, neck, trochanter, intertrochanter and shaft. Bottom center: bone compartments: cortical (blue), subcortical (yellow) and trabecular (red). Bottom right: oblique cut perpendicular to the neck axis separating the neck into quadrants

Full size image

Recent innovations for spine and hip QCT analyses were fostered by the interest in better understanding the effects of different pharmacological interventions separately on cortical, subcortical, and trabecular compartments as well in different VOIs, such as the femoral head and neck, trochanter, and intertrochanter in the proximal femur or sub VOIs within the vertebrae [21] (Fig. 1). QCT results have been reported from clinical trials of bisphosphonates [22, 23, 24•, 25–30], PTH [22, 24•, 26–29, 31], raloxifene [32], ronacaleret [24•], denosumab [33], and cathepsin K inhibitors [34]. Almost all of these studies were analyzed by software developed at the University of California (San Francisco, CA), the MIAF software from the University of Erlangen (Erlangen, Germany) or the commercially available and FDA approved QCTPro package (Mindways Inc). In addition to the hip and spine, the radius is also available to evaluation using regular CT equipment [35].

An important aspect in the interpretation of BMD results obtained from QCT is to realize that percentage changes depend on the baseline value. For example an absolute change of 10 mg/cm3 causes a relative change of 10 % in trabecular BMD (assuming an initial value of 100 mg/cm3) but only a 1.25 % increase of cortical BMD (assuming an initial value of 800 mg/cm3). Accordingly, in a recent cross-sectional QCT investigation of age related decreases in BMD a 2- to 5-fold higher percentage loss in trabecular vs cortical bone at all sites in the femur was reported, although absolute losses in trabecular and cortical bone were fairly similar [36]. The physiological, clinical, and biomechanical implications of this discrepancy, ie, the question whether an absolute or percent change is more relevant and for what, remains to be addressed.

Measuring Macrostructure

As QCT is a true 3D imaging modality it can provide structural information beyond BMD and BMC. Of particular renewed interest is the cortex, for example the local measurement of cortical thickness, density, and porosity. With MIAF as well as with the Bone Investigational Toolkit (BIT), an add-on to the QCTPro software, parameters such as integral, trabecular, and cortical cross-sectional areas and cross-sectional moments of inertia as well as endosteal and periosteal circumferences can be measured in slices perpendicular to the femoral neck axis [31, 37, 38]. These parameters can be applied to measures associated with bone strength such as section moduli and buckling ratios.

A principal limitation to accurately assessing the cortex is the limited spatial resolution of clinical CT equipment, of about 0.5 mm. With typical slice thicknesses of 1 mm and in plane pixel sizes of 0.3–0.7 mm the spatial resolution of currently used QCT protocols is usually not isotropic. The limitation in spatial resolution poses a problem in the spine, where the cortex is usually thinner than 0.5 mm as well as for sections of the femoral neck. Various segmentation strategies have been developed but the accuracy of cortical thickness, volume, and BMD measurements remains limited. Based on earlier theoretical results [39, 40] recently a new advanced cortical segmentation method has been proposed for the hip [41•, 42]. Accuracy errors of lower than 5 % for cortical thicknesses down to 0.3–0.4 mm were achieved based on a validation with femoral specimens scanned with hr-pQCT. This new algorithm requires the input of the “true” cortical density, which is measured in the compact bone of the shaft below the lesser trochanter but may differ at the site of the cortical measurement.

Cortical mass and thickness of the hip can now also be elegantly visualized as a topographical map. Using this technique significant cortical thickness reductions in specific anatomical regions and quadrants of the femoral neck and trochanter were identified, corresponding to the areas where the fracture initiated in fractured patients [43]. Using MIAF on QCT hip images, cortical BMD was reported to decrease in the superior quadrants of the neck 2- to 3-fold faster than observed in the inferior quadrants.

While cortical measures are increasingly being reported, trabecular dimensions prevent their accurate structural evaluation from clinical CT scans obtained with commonly used and accepted imaging protocols. Using increased radiation exposure and reducing slice thickness and pixel spacing, a higher resolution in the spine can be achieved. In combination with sophisticated image processing methods this technique (hr-CT) has been applied in vitro to assess fracture risk [44]. In vivo studies were reported in women to evaluate the effects of teriparatide [45] and in men with glucocorticoid-induced osteoporosis [46•, 47]. A comparison with hr-pQCT scans in excised vertebrae using a refined segmentation showed a high correlation of trabecular distance (r 2 = 0.98) [48] although other parameters were not reported. Further validation of hr-CT with μCT data and investigation of the effect of image noise appears crucial to fully understand the limitation of the spatial resolution of CT spine images on the accuracy of segmentation of individual trabeculae.

It is important to note that in in vitro studies often a better image quality is achieved than in vivo studies and therefore results of trabecular structure or texture may not easily be transferable. More work is required to better relate the cortical and trabecular geometric variables discussed in this section to fracture outcomes in untreated and treated patients and to bring these improved techniques to the clinical setting.

Measuring Microstructure

Bone microstructure typically refers to histomorphometric parameters originally obtained from 2D stained sections a few microns thick. This 2D approach has largely been replaced by imaging bone biopsy samples with μCT techniques offering isotropic 3D spatial resolution in the order of 10 μm. This resolution is however not achievable in vivo in humans due to the radiation exposure required. Currently in the distal forearm and tibia a high-resolution peripheral QCT (hr-pQCT) technique has been implemented on the XtremeCT scanner (Scanco, Brütisellen, Switzerland). The scanner provides 3D images with an isotropic voxel size of 41 μm or 82 μm, the latter resulting in isotropic spatial resolution of about 130–150 μm [49•]. With this technique the trabecular architecture can be analyzed resulting in parameters such as BV/TV, Tb.Th, Tb.Sp, and Tb.N in addition to BMD. Also cortical parameters such as BMD, thickness and porosity are calculated [50]. In the field of osteoporosis the technique has been used to predict or discriminate fractures [19, 51–57, 58•], and to assess age [59] and intervention related changes on peripheral bone architecture and BMD [60–68] as well as to investigate bone architecture under various other pathological conditions.

Data on hr-pQCT methods for fracture discrimination are still limited and most publications are based on the Strambo [51, 56] or Ofely studies [52, 55]. For many structural and density hr-pQCT parameters, age adjusted Odds ratios for discriminating fractures were in the range of 1.5–2. However, the contribution of structural parameters of the trabecular compartment became often insignificant after an additional DXA BMD adjustment. In contrast cortical thickness contributed independently of BMD in particular to the discrimination of vertebral fractures [51, 52, 56]. Interestingly, most intervention studies in addition to BMD emphasized hr-pQCT results for cortical thickness and density, and more recently for cortical porosity. In particular, in the radius effects on trabecular architecture parameters often were either not significant [60, 61, 65, 68], added little information to BMD [63] or were difficult to interpret [66].

There are multiple, potentially additive reasons for this observation: first the standard XtremeCT analysis protocol derives BV/TV from the trabecular bone density and an assumed compact bone density of 1200 mg HA/cm3. Therefore BV/TV is highly correlated with trabecular BMD. A fixed, global threshold is used to extract the trabecular structure for direct measurement of trabecular number and then Tb.Th and Tb.Sp are calculated using plate-model assumptions [69, 70]. Thus, changes in mineralization are neglected in the calculation of Tb.Th and Tb.Sp. Advanced segmentation and analysis techniques for the trabecular structure have been researched for hr-pQCT [69, 71–73] but remain largely experimental at present. Of course the underlying problem is the limited spatial resolution, which still causes significant accuracy errors in structural parameters [74•]. Least affected were BV/TV, Tb.N, and Ct.Th, whereas Tb.Th, Ct.Po, and Ct.Po.Dm were affected most. For these variables the correlation between hr-pQCT and μCT measurements, which served as a gold standard, were low or lacked significance.

Another reason is frequent motion artifacts caused by the long scan time of approximately 3 minutes of the devices currently used for the hr-pQCT measurements [75–78]. While more distinct artifacts can be easily recognized and the scan repeated, more subtle motion will cause a blurring of the trabecular network in the reconstructed image and therefore reduce the accuracy of the structural parameters without impacting on BMD results.

Regardless of the limitations, hr-pQCT has already contributed significantly to advance the understanding and interest in bone imaging. This is particularly true for cortical bone. Aging is associated with an increase in total outer bone volume, caused by periosteal apposition, and with cortical thinning and decrease in cortical BMD [79, 80]. The cortical BMD loss is primarily related to an increase in cortical porosity, which cannot be quantified by standard QCT or pQCT imaging due to spatial resolution limits. Cortical porosity increases from the periosteal to the endosteal surface [81] where subendosteal cavitation and conversion of the inner third of the cortex leads to a trabecular like structure [82], also termed trabecularization [83–85] of cortical bone.

Cortical porosity is now a standard parameter of the hr-pQCT analysis [86], however pores much smaller than the spatial resolution of the hr-pQCT scanner may not be properly accounted for due to partial volume artifacts. This is probably one reason why the correlation of cortical porosity results between hr-pQCT and μCT measurements was weak [74•]. This correlation and the accuracy of cortical porosity measurements has been improved by a recent technique that decreases the effect of partial volume artifacts on cortical porosity measurements by specific assumptions on cortical BMD [87•]. The problem of partial volume artifacts will also be much reduced by a separate analysis of the trabecularized part of the cortex where pore size is largely increased. Two recent publications have addressed the importance of such a separate analysis [87•, 88]. The added value of determining porosity over a plain cortical BMD measurement is an area of active research.

Measuring Strength

Strength of a bone can be experimentally determined only in a mechanical test, ie, ex-vivo, where typically the ultimate force necessary to fracture the bone is measured. In vivo we are restricted to an estimation of failure loads using FEA. The complicated problem of calculating deformations and stresses within a bone caused by external loads that may eventually cause fracture is addressed by creating a mesh of elements with known material properties for which equilibrium of forces and moments is enforced at both local and global scales (Fig. 2).

Fig. 2
figure 2

μFE model of the distal radius (top) and homogenized FE model of the vertebral body (bottom) both under axial compression

Full size image

In the bone field, typically μFEA models are distinguished from homogenized FEA models. The former ones require high spatial resolution images so that the elements of the mesh contain bone material only. A segmentation step to separate the individual trabeculae as well as the pores in the compact bone precedes the formation of the mesh model of the bone of interest. Appropriate imaging modalities are μCT, used for specimen work and hr-pQCT [56, 57, 61, 89–94] as well as hr-MRI [95, 96] that can be applied in vivo at the distal forearm and tibia. Dedicated FEA analysis software is available from the manufacturer for the XtremeCT scanner but many research groups use their own proprietary developments.

Thanks to the high resolution of pQCT images available in the past decade, experimental validation of the μFE models progressed to high levels with refined ex vivo biomechanical tests mimicking Colles’ fracture [89, 92, 94, 97]. Interestingly, the outstanding prediction of ultimate load of the distal radius by μFEA was found to be only slightly superior to the one of BMC [98].

While the radius and tibia have enabled important advances in this area, the main fracture sites are the spine and hip, where such a high resolution cannot be achieved. For these skeletal locations homogenized FE models are applied to QCT images obtained from whole body clinical CT-scanners (Fig. 2). A mesh is applied to the entire vertebral body or proximal femur resulting in individual elements with a size in the mm range but higher than the voxel dimensions of the CT dataset containing mineralized bone as well as bone marrow. The term homogenization designates the averaging process to determine the apparent material properties of that mixture of bone and marrow, which are essential input parameters for FEA. Bone is an inhomogeneous, anisotropic and elastic material that undergoes simultaneous plastic deformation and loss of stiffness due to initiation and growth of micro-cracks when overloaded [99, 100].

However, in single energy CT images the only available information of a given voxel is its absorption coefficient or, after proper calibration, its apparent BMD value. The BMD values are linearly or more often non-linearly mapped to the Young’s modulus for bone stiffness and to the yield criterion for failure load [101, 102, 103••, 104–107], which is the main outcome variable. Other parameters such as stress, strain, or damage (loss of stiffness due to microcracks) distribution maps also can be obtained from FEA.

In recent years homogenized FEA of the spine and hip has been applied in epidemiological studies to determine strength or to predict spine [16••, 58•, 108] or hip [109, 110] fractures or to determine changes of bone strength with age [111] or with different treatment [23, 26–28, 34, 112–115]. The accuracy of FEA has been extensively validated using biomechanical tests where the applied forces can be well predicted in the FE model. However, in vivo, the magnitude and direction of external forces, in particular those that occur during a specific fall, can only be estimated [116, 117] and impact the resulting fracture risk prediction. To address this limitation, FEA is often performed on multiple loading conditions such as axial compression, anterior bending, and torsion scenarios in the vertebrae, and for stance and fall scenarios in the proximal femur. Preliminary results indicate that the vertebral body stiffnesses along the different loading scenarios are highly correlated [118]

A performance comparison between FEA and BMD results is difficult. In a recent in vitro study of vertebral sections [119••] FEA was superior to BMD as measured by DXA or QCT but not to BMC in predicting ultimate load. However ultimate load is not size adjusted and larger bones can withstand larger load before they fracture. In the same study, FEA performed only slightly better than BMD as measured by QCT for prediction of the size adjusted parameter apparent ultimate stress. In another in vitro study ultimate load was also better predicted by FEA than by QCT BMD but there BMD was only measured in the trabecular compartment whereas the mechanical tests and the FE calculations were performed on complete vertebral bodies [120]. Femoral strength also was found to be better predicted by FEA than QCT and DXA [121] but in this early study FEA applied to the total proximal femur was compared with QCT and DXA measurements from the neck. In a more recent work, the predictions of femoral strength in stance configuration by FEA were not significantly better than those by BMC measured from QCT [103••].

In MrOs, a cohort study in elderly males, in a retrospective subset analysis ultimate load by FEA in the hip improved fracture prediction, but only marginally compared with DXA [109]. Interestingly in the AGES study male-female differences in the association between incident hip fracture and proximal femoral strength were found [110] although BMD of the hip as measured by DXA predicts hip fractures equally well in both sexes [122]. In MrOs, in the spine, hazard ratios and areas under the curve (AUC) for ultimate load and integral BMD of the total vertebral body were significantly higher than DXA areal BMD [16••]. QCT BMD results were not reported. Similar results in the spine were reported for Japanese women [25], although here no AUC values were given. This is a limitation in particular if reported confidence intervals for hazard or odds ratios from FEA calculations are considerably larger than those for DXA. Even large differences in hazard or odds ratios among techniques may then not be significant. Notwithstanding these limitations there is growing evidence that for fracture prediction at the spine QCT and FEA are superior to DXA.

Several reports on FEA from therapeutic interventions are available. In general the largest treatment-related differences between DXA and FEA results were observed in the spine emphasizing limitations of the projectional DXA technique [23, 26, 34, 47, 112, 113]. Smaller numerical differences were observed at the hip [23, 27, 34, 115] but the significance of the difference was rarely reported. The comparison between FEA and QCT results is even more difficult, in particular, as often QCT and FEA analyses VOIs do not match.

The unique attribute of FEA is the integration of bone geometry and density distribution to calculate bone strength for various load cases. Accordingly FEA can provide additional valuable information. FEA is a rapidly evolving technique and while most FEA software for bone analysis is university based and experimental, one such methodology (ON Diagnostics, Berkley, CA) recently received FDA approval for the use in identification of patients at risk for fracture and follow-up the effect of therapies.

Outlook and Summary

In less than 3 decades since the first report of the QCT methods to measure BMD of the spine and radius, many substantial technical and methodological improvements have been introduced and applied. The CT based techniques reviewed in this article have enabled improved fracture risk prediction, clarification of the pathophysiology of skeletal disease, characterization of the skeletal response to therapy, and enhanced assessment of important biomechanical relationships. They are used extensively in research and increasingly in clinical trials and, over the last decade, have significantly improved our understanding of the organ bone. DXA remains the gold standard in practice, despite its known limitations for monitoring pharmaceutical treatment. However, CT based techniques provide important complementary information. While much remains to be done we can now elegantly visualize and quantitate bone biology in-vivo, thus benefitting science, physician and patient.